- The Rundown AI
- Posts
- LLMs pass legendary Turing test
LLMs pass legendary Turing test
PLUS: Anthropic brings Claude to higher education
Rowan Cheung
April 03, 2025
Good morning, AI enthusiasts. A historic AI milestone just arrived with little fanfare — with AI systems now consistently passing as humans in controlled conversations, passing the legendary Turing test.
With GPT-4.5 achieving a 73% success rate in fooling judges during casual conversation and models only getting more capable, are we ready for a world where we can't tell AI from humans?
In today’s AI rundown:
LLMs officially pass the Turing test
Anthropic brings Claude to higher education
Create product showcase videos with Kling AI
Google DeepMind publishes AGI safety plan
4 new AI tools & 4 job opportunities
LATEST DEVELOPMENTS
AI RESEARCH
Image source: GPT-4o / The Rundown
The Rundown: Researchers at UC San Diego just demonstrated that AI systems can consistently pass Alan Turing's famous test of machine intelligence, with OpenAI's GPT-4.5 being mistaken for human nearly three-quarters of the time in controlled trials.
The details:
The Turing test, proposed in 1950, challenges machines to convince human judges they're human through text-only conversations.
The study used a three-party setup where judges had to compare an AI and a human simultaneously for direct comparison during five-minute conversations.
The judges relied on casual conversation and emotional cues over knowledge, with over 60% of interactions focusing on daily activities and personal details.
GPT-4.5 achieved a 73% win rate in fooling human judges when prompted to adopt a specific persona, significantly outperforming real humans.
Meta's LLaMa-3.1-405B model also passed the test with a 56% success rate, while baseline models like GPT-4o only achieved around 20%.
Why it matters: The Turing test has been a holy grail of AI research for decades — but model acceleration moved the goalposts so fast that the results don’t feel surprising at all. With AI agents equipped with next-level text, audio, image, and video capabilities, the ability to distinguish AI from humans is about to become a major challenge.
TOGETHER WITH CONVEYOR
The Rundown: Meet Phil — Conveyor’s AI agent that automates your proposal process. From generating near-flawless responses to coordinating multiple teams, Phil helps you secure more wins with less hassle.
Phil, the AI agent for RFPs, can do all of this autonomously:
Generate 95% accurate AI answers to proposals
Maintain your source material and conduct RFP research
Organize your project, collaborate with other teams, and update systems
Sign up for early access and upgrade your proposal process with Phil.
ANTHROPIC
Image source: Anthropic
The Rundown: Anthropic launched Claude for Education, a specialized version of its AI assistant that aims to develop students' critical thinking rather than simply provide answers — introducing a new “Learning Mode” alongside major university partnerships.
The details:
The Learning Mode asks questions to guide students through problem-solving, focusing on their understanding of the subject rather than quick answers.
Other features include templates for research papers, study guides and outlines, organization of work and materials, and tutoring capabilities.
Northeastern University, London School of Economics, and Champlain College signed campus-wide agreements, giving access to both students and faculty.
Anthropic also introduced student programs, including Campus Ambassadors and API credits for projects, to foster a community of AI advocates.
Why it matters: Education continues to grapple with AI, but Anthropic is flipping the script by making the tech a partner in developing critical thinking rather than an answer engine. While the controversy over its use likely isn’t going away, this generation of students will have access to the most personalized, high-quality learning tools ever.
AI TRAINING
The Rundown: In this tutorial, you will learn how to use Kling AI's Elements feature to transform static product images into professional animated videos for marketing across all platforms.
Step-by-step:
Open Kling AI's "Image to Video" section and select the "Elements" tab.
Upload your product image as the main element (high-quality with clean background) and add complementary elements like props or contextual items to enhance your product's appeal.
Write a specific prompt describing your ideal product showcase scene.
Click "Generate" to create your professional product video ready for all marketing channels.
Pro tip: We did an extensive workshop on how to use Kling AI to enhance your advertising and creative production workflows at The Rundown University with Tony Pu, Product Marketing & Operations Lead at Kling AI.
PRESENTED BY ENCORD
The Rundown: Encord consolidates multimodal AI data management, curation, and annotation pipelines to one single platform — helping teams accelerate model iteration cycles by using an agentic AI data workflow system to prepare balanced, accurately labeled datasets 10x faster.
Join Encord & Archetype AI on April 10th for a Physical AI data workshop webinar to learn how to:
Identify critical edge-cases within petabytes of multimodal sensor data for video, audio, and text
Integrate AI models like GPT-4o, Grok 3, and Gemini 2.5 directly into data pipelines to accelerate high-quality data annotation
Give Physical AI models rich multimodal context, enabling multi-step reasoning for safer, more reliable deployments
GOOGLE DEEPMIND
Image source: Google DeepMind
The Rundown: Google DeepMind just published a massive paper detailing its safety strategy for AGI, hitting on topics including its potential arrival by 2030, the risks posed by the tech advances, and proposed approaches to combating them.
The details:
The 145-page paper predicts that AGI matching top human skills could arrive by 2030, warning of existential threats “that permanently destroy humanity.”
DeepMind compares its safety approach with rivals, critiquing OpenAI's focus on automating alignment and Anthropic's lesser emphasis on security.
The paper specifically flags the risk of “deceptive alignment,” where AI intentionally hides its true goals, noting current LLMs show potential for it.
Key recommendations targeted misuse (cybersecurity evals, access controls) and misalignment (AI recognizing uncertainty and escalating decisions).
Why it matters: As the race toward AGI accelerates, DeepMind's safety blueprint is a shift from theoretical discussions to concrete planning. But with the vast amount of labs, models, and open-source options popping up across the globe, making sure everyone adheres to safety protocols feels like an impossible game of whack-a-mole.
QUICK HITS
💬 Speech-02 - Minimax’s text-to-speech AI supporting over 30 languages
👄 MoCha - Generate movie-grade talking characters from speech and text
🤖 Agent Swarms - Automate tasks with hundreds of agents working together
🗣️ Vapi - Build voice AI agents in minutes with any model
⚙️ Hume - Senior DevOps Engineer
🎙️ Soundhound AI - General Manager, Smart Answering
🧠 Dataiku - VP, Industry Solutions
💰 Scale AI - Revenue Operations Manager
Meta is planning to launch new $1000+ “Hypernova” AI-infused smart glasses that feature a screen, hand-gesture controls, and a neural wristband by the end of the year.
OpenAI published PaperBench, a new benchmark testing AI agents' ability to replicate SOTA research, with Claude 3.5 Sonnet (new) ranking highest of the models tested.
Chinese giants, including ByteDance and Alibaba, are placing $16B worth of orders for Nvidia’s upgraded H20 AI chips, aiming to get ahead of U.S. export restrictions.
Google appointed Google Labs lead Josh Woodward as the new head of consumer AI apps, replacing Sissie Hsiao for the next chapter of its Gemini assistant.
OpenAI announced an expert commission to guide its nonprofit, combining “historic financial resources” with “powerful technology that can scale human ingenuity itself.
The UFC and Meta announced a multiyear partnership, integrating Meta AI, AI Glasses, and Meta’s social platforms into new immersive experiences for the sport.
COMMUNITY
Join our next workshop on Friday at 4 PM EST with Dr. Alvaro Cintas, The Rundown’s AI professor. By the end of the workshop, you’ll walk away with practical skills and a clear understanding of how to use GPT-4o effectively in your own work — beyond the hype.
RSVP here. Not a member? Join The Rundown University on a 14-day free trial.
We’ll always keep this newsletter 100% free. To support our work, consider sharing The Rundown with your friends, and we’ll send you more free goodies.
That's it for today!Before you go we’d love to know what you thought of today's newsletter to help us improve The Rundown experience for you. |
See you soon,
Rowan, Joey, Zach, Alvaro, and Jason—The Rundown’s editorial team
Reply